Membrane permeability differentiation at the lipid divide

One of the deepest branches in the tree of life separates the Archaea from the Bacteria. These prokaryotic groups have distinct cellular systems including fundamentally different phospholipid membrane bilayers. This dichotomy has been termed the lipid divide and possibly bestows different biophysical and biochemical characteristics on each cell type. Classic experiments suggest that bacterial membranes (formed from lipids extracted from Escherichia coli for example) show permeability to key metabolites comparable to archaeal membranes (formed from lipids extracted from Halobacterium salinarum), yet systematic analyses based on direct measurements of membrane permeability are absent. Here we develop a new approach for assessing the membrane permeability of ~10 μm unilamellar vesicles, consisting of an aqueous medium enclosed by a single lipid bilayer. Comparing the permeability of eighteen metabolites demonstrates that diether glycerol-1-phosphate lipids with methyl branches, often the most abundant membrane lipids of known archaea, are permeable to a wide range of compounds useful for core metabolic networks, including amino acids, sugars, and nucleobases. Permeability is significantly lower in diester glycerol-3-phosphate lipids without methyl branches, the common building block of bacterial membranes. To identify the membrane characteristics that determine permeability we use this experimental platform to test a variety of lipid forms bearing a diversity of intermediate characteristics. We found that increased membrane permeability is dependent on both the methyl branches present on the archaeal phospholipid tails and the ether bond between the tails and the head group. These permeability differences must have had profound effects on the cell physiology and proteome evolution of early prokaryotic forms. To explore this further, we compare the abundance and distribution of transmembrane transporter-encoding protein families present on genomes sampled from across the prokaryotic tree of life. Archaea have a reduced repertoire of transporter gene families, consistent with increased membrane permeation. These results demonstrate that the lipid divide demarcates a clear difference in permeability function with implications for understanding some of the earliest transitions in cell evolution.


INTRODUCTION
Membranes are the boundaries that define cells. At some stage in early evolution, a molecular system must have become encapsulated within a lipid membrane [1] with functions that allowed metabolites to permeate the membrane and provide substrates for the molecular system to function. How metabolites cross membranes is therefore a key factor for understanding early evolution. Several key steps later, the prokaryotes emerged, cellular forms which include the Bacteria (once called Eubacteria) and the Archaea (once called Archaebacteria) [2][3][4][5][6]. The placement of the root of the prokaryotes is debated [7][8][9], making it difficult to understand the biology of the last universal common ancestor (LUCA). However, the Archaea/Bacteria bifurcation, which could represent LUCA, marks multiple important differences in cell biology. The consequences of these differences are key to understanding how the Archaea and Bacteria emerged.
The main function of cell membranes is to generate a semi-permeable solute barrier that allows for the development of chemical and proton gradients that drive the biochemistry and energetics of life [19]. The core difference between the G3P diester lipids of Bacteria and G1P diether lipids with methyl branches of Archaea could have a profound effect on the primary function of the cell membrane. In vitro studies have identified important biological implications of the physico-chemical properties of phospholipid membranes by using liposomes [20][21][22][23][24][25][26][27][28][29]. Liposomes are approximately spherical synthetic lipid bilayer membranes with a typical diameter of 100 nm that enclose an internal aqueous phase. A previous study using liposomes suggested that key metabolites from the environment could permeate prebiotically plausible membranes in the absence of transport machinery such as transmembrane transporter proteins [30]. However, this foundational work did not contrast the permeability of core metabolites across archaeal and bacterial-type membranes, but rather used mixtures of simple prebiotically plausible lipids, such as fatty acids, fatty alcohols and monoglycerides. An additional study has shown that, in the absence of counter-ions, liposomes made of diether lipids extracted from archaea (e.g. from Halobacterium salinarum) display lower permeability to protons compared to liposomes made of lipids extracted from bacteria (e.g. from Escherichia coli) [16]. This feature was confirmed using independent approaches [28,31,32]; the variation in proton permeability being likely due to the difference in the ether versus ester head-tail bond [33]. This differential permeability constitutes an important trait for understanding the differences in energetic functions of the Archaea and Bacteria [19]. It was also shown that bacterial and diether archaeal lipid membranes display similar permeabilities to glycerol, urea and ammonia [16]. However, these permeability traits were measured at high extracellular metabolite concentration (i.e. 200 mM) using indirect spectroscopic techniques, which average over a large number of liposomes and associated impurities (e.g. multi-lamellar liposomes and lipid aggregates). As such there are a number of potential limitations of these results: i) they are based on ensemble measurements across variant lipid architectural forms, ii) they do not investigate permeability characteristics of cellular metabolites at biologically/ecologically realistic substrate concentrations, and iii) they investigate a limited number of metabolites. Collectively, this means that there is a limited understanding of the implications of membrane chemistry variation on metabolite permeability in biologically relevant contexts, particularly given recent evidence suggesting that living cells can obtain nutrients from the environment in a transporter proteinindependent fashion [34].
Here we test the hypothesis that core archaeal and bacterial type lipid membranes have fundamentally different permeability traits directly implying that the ancient Archaea/Bacteria bifurcation would also encompass a distinct change in metabolite permeability. Such a difference would have profound effects for the evolution of membrane transporter repertoires, intracellular metabolic networks, and associated cellular ecologies. We present a new approach for the study of membrane permeability based on microfluidic manipulation of unilamellar vesicles composed of a single phospholipid bilayer of archaeal or bacterial membrane mimics. Our results show that the lipid divide demarcates a dichotomy in membrane permeability characteristics. In contrast to previous ideas, membranes composed of archaeal core phospholipids display elevated permeability to many compounds key for core metabolic functions. Using phylogenomic approaches, we also demonstrate that this functional difference correlates with variations in the evolution of the transporter protein encoding gene repertoire.

Microfluidic screening to explore membrane permeability characteristics
Here we report a system to enable the capture and individual placement of unilamellar vesicles obtained via electroformation of synthetic lipids (see Methods) in multiple parallel arrays of tens of vesicles using microfluidics ( Figure 1). This approach enabled us to precisely control the chemostatic fluid environment of the vesicles [35]. The diameter of these vesicles is in the range of 5-to 15-µm. Although such diameter range is large for prokaryotic cell sizes, these dimensions were chosen to aid imaging and manipulation. In order to measure metabolite permeation into the unilamellar vesicles, we loaded the vesicles with a neutral pH buffer and carboxyfluorescein (CF), a fluorescent dye which allows direct assessment of molecular uptake by variations in fluorescent properties in response to changes in intra-vesicle metabolite concentration [16,36]. In fact, the introduction of metabolites in the vesicle reduces the self-quenching properties of carboxyfluorescein, resulting in increased vesicle fluorescence. Therefore, a relative increase in intra-vesicle fluorescence indicates membrane permeability to the target metabolite when delivered via continuous flow through the microfluidic device (Figure 1 and Methods). Individual metabolites were delivered into the extra-vesicle environment (also containing a neutral pH buffer), while imaging the changes in fluorescence levels of multiple individually trapped vesicles (Figure 1 and Methods). By using this experimental approach, we conducted parallel controlled experiments exploring how cellular metabolites can cross membranes of different phospholipid chemical composition using changes in CF fluorescence as a reporter for membrane permeability to metabolites.
We first studied the permeability of single vesicles composed of synthetic lipids with isoprenoid chains containing methyl branches bonded to a glycerol-1-phosphate (G1P) backbone via ether bonds (archaeal-like membrane phospholipids -abbreviated here as 'archaeal 4ME diether G1PC') (lipid 1 in Table S1) or synthetic lipids with fatty acids bonded to a glycerol-3-phosphate (G3P) backbone via ester bonds (bacterial-like membrane phospholipids -ternary lipid mixture abbreviated here as 'bacterial diester G3PE-PG-CA', lipid 2 in Tables S1) to eighteen small metabolites (Table S2). Notably, synthetic G1P diether lipids with methyl branches are not commercially available, to our knowledge, and were therefore synthesised de novo for the purpose of this study (see Methods). We chose metabolites with different molecular weight, hydrophobicity, and number of rotatable bonds (see Table S2).
Our single-vesicle measurements revealed heterogeneity in the permeability of each synthetic lipid type to each metabolite, with some vesicles of each lipid type displaying a decrease in intracellular fluorescence during the delivery of each metabolite while other vesicles displayed an increase in intracellular fluorescence (temporal dependence of single-vesicle fluorescence for archaeal 4ME diether G1PC lipids and bacterial diester G3PE-PG-CA lipids are reported with dashed magenta lines and dashed-dotted blue lines in Figure S1). For example, the coefficient of variations in the permeability of archaeal 4ME diether G1PC vesicles and of bacterial diester G3PE-PG-CA vesicles to aspartic acid, glyceraldehyde and adenine, were 118% and 103%, 60% and 198%, 76% and 108% (Figure S1D, S1G and S1Q, respectively), in accordance with the proposition that lipid membranes are a system with heterogeneous functions [37,[39][40][41][42][43].
Therefore, in order to compare the permeability of different synthetic lipid types to the same metabolite, we carried out Mann-Whitney two-tailed statistical comparisons between the distributions of intra-vesicle fluorescence after 3 minutes of delivery of each metabolite for each synthetic lipid type. Temporal dependence of fluorescence distribution means and standard deviations are reported in Figure 2 (magenta triangles and dashed lines for archaeal 4ME diether G1PC vesicles, blue squares and dashed-dotted lines for bacterial diester G3PE-PG-CA vesicles) together with single-vesicle fluorescence distributions and Mann-Whitney two-tailed statistical comparisons at t=3 (further details about Mann-Whitney two-tailed statistical comparisons are reported in Supplementary file 1).
We found that archaeal 4ME diether G1PC vesicles were significantly more permeable to the amino acids glycine, alanine, leucine, aspartic acid, tryptophan and glutamine compared to bacterial diester G3PE-PG-CA vesicles (Figure 2A-2F, Figure S1A-S1F, Supplementary file 1). Consistent with the amino acid findings, archaeal 4ME diether G1PC vesicles were also more permeable to the sugar's glyceraldehyde, glycerol, deoxyribose, ribose and arabinose compared to bacterial diester G3PE-PG-CA vesicles ( Figure 2G-2K, Figure S1G-S1K, Supplementary file 1), whereas we did not measure a significant difference in the permeability to dihydroxyacetone ( Figure 2L, Figure S1L, and Supplementary file 1). Interestingly, the difference in permeability was strongly distinct for three large sugar types, deoxyribose, ribose, and arabinose, the two former sugars including primary constituents of the hereditary materials DNA and RNA, respectively.
Finally, archaeal 4ME diether G1PC vesicles were also more permeable to the amide urea, and the nucleobases cytosine, uracil, guanine and adenine compared to bacterial diester G3PE-PG-CA vesicles ( Figure   2M-2Q, Figure S1M-S1Q, Supplementary file 1). These data suggest that important nitrogen sources and components of DNA and RNA can permeate archaeal 4ME diether G1PC vesicles. The phosphonate 2aminoethyl phosphonic acid ( Figure 2R, Figure S1R, Supplementary file 1) showed no significant difference in permeability characteristics between the two types of vesicles. We also attempted to measure permeabilities to the nucleotide adenosine monophosphate, however, during pilot studies the vesicles burst during delivery of this chemical and so we did not test this substrate further.
Pooling together the data for the eighteen metabolites investigated, we did not find a significant correlation between permeability and lipid molecular weight or hydrophobicity (Pearson coefficient r = -0.22, 0.36, respectively, for archaeal 4ME diether G1PC vesicles; r = 0.09, 0.24, respectively, for bacterial diester G3PE-PG-CA vesicles; none significant at p = 0.05; Table S2). These data suggest that the metabolic selectivity of these membrane mimics is complex and does not rely solely on basic molecular properties such as molecular weight or hydrophobicity. Taken together with previous findings demonstrating that solute hydrophobicity correlates poorly with permeability coefficients of eukaryotic fatty acid or phospholipid membranes [36], our data corroborate the hypothesis that subtle variations in the metabolite atomic structure could contribute to differences in membrane permeability [36][37][38]. In fact, we found a significant correlation between permeability and rotatable bond number for archaeal 4ME diether G1PC vesicles (Pearson coefficient r = -0.49, *) but not for bacterial diester G3PE-PG-CA vesicles (r = -0.32, non-significant at p = 0.05).
The striking difference in CF fluorescence between archaeal 4ME diether G1PC and bacterial diester G3PE-PG-CA vesicles was confirmed when we extended the duration of our permeability experiments from 3 to 6 minutes ( Figure S2). Moreover, the dimensions of each vesicle did not significantly change during metabolite delivery ( Figure S3), confirming that vesicle deformation did not occur in our experiments and did not play a role in the observed differences in permeability traits between archaeal 4ME diether G1PC vesicles and bacterial diester G3PE-PG-CA vesicles. Such differences were also not attributable to differential metabolite accumulation within the lipid bilayers, since: i) CF has very low affinity for the lipid hydrophobic chains [16], so interactions with substrates within the membrane are unlikely; ii) CF fluorescence intensity was uniform across whole vesicles for both lipid type and the full range of metabolites investigated, consistent with previous data obtained on fatty acid and phospholipid liposomes [16,36]. Our data do not allow us to infer a detailed kinetics of the permeation of each metabolite and could be complemented via pulse-chase experiments requiring faster fluidic exchanges. Therefore, we did not attempt to extract absolute kinetic parameters, such as the permeability coefficient, but centred this current work on directly comparing changes in CF fluorescence (as a proxy for permeability) between archaeal 4ME diether G1PC and bacterial diester G3PE-PG-CA vesicles. However, we were able to observe differences in terms of both the uptake onset and slope for different metabolites. For example, glycine, ribose and uracil displayed a steep uptake during the first minute of their delivery to archaeal 4ME diether G1PC vesicles (Figure 2A, 2J and 2O, respectively), whereas the uptake of aspartic acid, glutamine and dihydroxyacetone started only after the first minute of their delivery to archaeal 4ME diether G1PC vesicles ( Figure 2D, 2F and 2L). These data demonstrate that metabolites are not passing across vesicles via puncture holes generated during vesicle formation but by genuine diffusion through the lipid bilayers in a metabolite-specific manner.
As discussed in the introduction, natural archaeal membranes are formed from heterogenous mixtures of lipids some with tetraether bipolar lipids (e.g. caldarchaeol) which act to directly connect the membrane bilayers, a function that is likely to increase the stiffness of the membrane and reduce permeability [44]. Such mixtures might have different properties than the homogenous membranes studied here. To explore this possibility, we attempted to use lipids extracted from Haloferax volcanii, predominantly containing diether lipids with head group derivatives of phosphatidylglycerol [45], and lipids extracted from Sulfolobus acidocaldarius, predominantly containing tetraether lipids with head group derivatives of phosphatidylhexose [46]. However, we could not obtain mechanically stable vesicles via electroformation for either of these lipid mixtures. It is therefore important to mention that our experiments do not reveal the permeability traits of extant prokaryotic membrane mixtures but rather identify the contrasting permeability traits of the common and core building blocks of the archaeal and bacterial membranes.
Finally, we wanted to rule out that the relatively lower permeability of bacterial diester G3PE-PG-CA vesicles could be due to interactions between different lipids within the ternary lipid mixture that we employed to mimic more closely bacterial membranes (lipid 2 in Table S1). To do so, we measured and contrasted permeability to urea, glycine, ribose, deoxyribose, glycerol and phosphonate in vesicles made of bacterial ternary-lipid mixtures (G3PE-PG-CA, lipid 2 in Table S1) and vesicles made of single lipids (G3PE, lipid 6 in Table S1) and found that these two different bacterial mimics displayed comparably low permeabilities to all the metabolites tested ( Figure S4).

Which archaeal lipid characteristics determine permeability traits?
To uncover the chemical determinants of archaeal membrane permeability, we employed vesicles made of a range of lipids with a mixture of archaeal and bacterial lipid characteristics. We tested the impact of the lipid chain branching, length, tail-head bond (ester/ether) and the G1P vs G3P backbone on membrane permeability. We performed these experimental tests using urea, glycine, ribose, deoxyribose, glycerol and phosphonate towards which archaeal 4ME diether G1PC vesicles and bacterial diester G3PE-PG-CA vesicles display different patterns of permeabilities ( Figure 2).
Firstly, we electroformed vesicles by using a lipid that carried a bacterial-like glycerol-3-phosphate (G3P) backbone but with archaeal like diether tail-head bond and isoprenoid chains containing methyl branches (4ME diether G3PC, lipid 3 in Table S1, green circles in Figure 3). The permeabilities of these hybrid vesicles to the six metabolites investigated were not significantly different to the permeabilities of archaeal vesicles (4ME diether G1PC, lipid 1 in Table S1, magenta upward triangles in Figure 3 and Supplementary file 1) but were significantly higher than the permeabilities measured for the bacterial vesicles (diester G3PE-PG-CA, lipid 2 in Table S1, blue squares in Figure 3, Kruskal-Wallis one-way analysis of variance statistical comparisons are reported in Supplementary file 2). These data suggested that the change from a G1P to a G3P backbone was not a key factor in determining membrane permeability.
Next, we tested a different hybrid lipid with the bacterial-like glycerol-3-phosphate (G3P) backbone but with archaeal-like diether tail-head bond and isoprenoid chains without methyl branches (Diether G3PC, lipid 4 in Table S1 and black diamonds in Figure 3). We found that in the absence of lipid chain branching, these hybrid vesicles displayed a statistically significant and consistently lower permeability compared to archaeal 4ME diether G1PC vesicles (magenta upward triangles in Figure 3 and Supplementary file 2). This striking difference is possibly due to increased membrane fluidity in the presence of methyl branched lipid chains [30,47]. In the absence of methyl branching, these hybrid diether G3PC vesicles displayed permeabilities to urea, glycine, deoxyribose or ribose that were comparable to the permeabilities measured for bacterial diester G3PE-PG-CA vesicles (lipid 2 in Table S1 and blue squares in Figure 3A, 3B, 3E and 3F, respectively, none significantly different at p = 0.05, Supplementary file 2). In contrast, permeabilities to glycerol and phosphonate were lower in the hybrid diether G3PC vesicles without methyl branching compared to bacterial diester G3PE-PG-CA vesicles ( Figure 3C and D, * and *, respectively, Supplementary file 2).
A third hybrid lipid with the bacterial-like glycerol-3-phosphate (G3P) backbone and a diester tail-head bond but with archaeal-like isoprenoid chains with the methyl branches was also compared in a similar manner (4ME Diester G3PC, lipid 5 in Table S1, brown downward triangles in Figure 3). In the absence of an ether bond (substituted with an ester bond), 4ME diester G3PC vesicles displayed significantly lower permeabilities to all six metabolites investigated compared to archaeal 4ME diether G1PC vesicles (magenta upward triangles in Figure 3 and Supplementary file 2), and exhibited permeabilities similar to bacterial diester G3PE-PG-CA vesicles (blue squares in Figure 3, none significantly different apart from permeability to ribose that was significantly lower in 4ME Diester G3PC vesicles, *, Supplementary file 2). Taken together these data demonstrate that increased membrane permeability can be achieved via the simultaneous use of an ether bond and methyl chain branching, both of which characterize the core lipids of archaea.
To follow on from this, we set out to determine whether permeability is affected by variations in the archaeal lipid head (lipid 7 in Table S1). However, despite attempting different electroformation protocols (see Table S4) we could not produce vesicles using this lipid, possibly because this lipid forms non-lamellar structures of either a cubic or hexagonal fashion [48]. We note, however, that previous studies suggest that lipid headgroup composition has only limited impact on the permeation of small molecules [29].
Next, we investigated how permeability varies in phospholipids according to chain length. Synthetic G1P lipids with methyl branches are not commercially available, so we focused on the study of G3P lipids without methyl branches and with variant chain lengths (lipids 4, 8, 9 in Table S1). These experiments provide no evidence for a significant correlation between lipid chain length (C=12, C=16 and C=18, i.e. lipids 8, 4 and 9 in Table S1) and permeability to any of the metabolites investigated ( Figure S5, Pearson correlation coefficient r=-0.58, 0.95, 0.52, -0.10, -0.69 and 0.30 for urea, glycine, glycerol, phosphonate, deoxyribose and ribose, respectively), corroborating previous findings on eukaryotic membrane mimics [49]. Significant differences in permeability to water and weak acids were previously observed only for large variations in chain length; an increase from 14 to 26 in the length of acyl chains of eukaryotic lipids led to a five-fold decrease in permeability [29]. Accordingly, our data show that a short chain length slightly favoured permeability to urea and deoxyribose ( Figure S5A and S5E, respectively). In contrast, a long chain length slightly favoured permeability to glycine and ribose ( Figure S5B and S5F, respectively), but overall, these effects were masked by vesicle-to-vesicle variation in permeability to these metabolites. Furthermore, we attempted to produce vesicles using lipids of other chain lengths (C=6 and C=14, i.e. lipids 10 and 11 in Table S1). However, these vesicles appeared to be mechanically unstable, possibly because their transition temperature is close to the temperature at which we carried out our membrane permeability assays. We note, these comparisons did not include variations in the number of methyl branches per chain so our experiments do not rule out the possibility that differences in the number of methyl branches may alter permeability characteristics.
Our next experiments demonstrated that decreasing bonding saturation (i.e. single bonds that were more likely present in prebiotic molecules [30] versus double bonds) along hybrid G3P diether phospholipids of fixed chain length (C=18, lipid 12 in Table S1) significantly decreased permeability to the small amide, urea, and, to a lesser extent, to the small amino acid, glycine, compared to bonding unsaturation (i.e. double bonds, lipid 11 in Table S1, Figure S6A and S6B, *** and *, respectively). However, bonding saturation did not have an impact on the permeability to glycerol, phosphonate, deoxyribose, or ribose ( Figure S6C, S6D, S6E and S6F, respectively, none significantly different), whereas previous studies using G3P diester phospholipids or fatty acid liposomes found increased permeability to glycerol [29] and ribose [30] liposomes made of unsaturated lipids.
Tetraether bonds, generating bipolar lipids (or caldarchaeol), or cyclopentane rings along the caldarchaeol chains could further affect the permeability traits in archaeal membrane mimics. Given the variance of these lipid forms across Archaea [50] and the observation that such variants are the minority constituents in some archaeal membranes [51,52], we suggest they are of lesser importance for understanding the evolution of core/ancestral permeability functions in ancestral cell forms carrying archaeal lipids. However, we do acknowledge that introduction of these lipid variations is likely to alter permeability [53] and is an important factor for determining the ecology of extant Archaea. The experimental platform presented is readily adaptable to investigate the effect of further chemistry variations, however, these synthetic lipids are not currently commercially available. Future work should explore the effect of lipid mixtures on permeability traits. Furthermore, we did not investigate the effect on permeability of membrane variants embedding an S-layer, a peptidoglycan layer, or an outer membrane such as those seen in diderm bacteria, because all these features are predicted to follow the archaeal/bacterial bifurcation and so represent secondary elaborations relative to the lipid divide. We also acknowledge that permeability traits can vary with changes in environmental conditions such as pH (which we kept constant at 7.4), temperature (which we kept constant at 22 °C) or salinity levels [54,55] and the differences we have observed are likely subject to variation in different environmental conditions. Although our experimental platform could be adapted to investigate permeability traits under variant environmental conditions, there are constraints as vesicles suffer damage in the presence of low or high pH and/or salinity. Moreover, lipids change state above or below the transition temperature.
Therefore, the experimental platform would need further development to physically stabilise vesicles, by using, for example, higher density media or by forming vesicles on physical support structures. However, such experiments would tell us much about the conditions in which cellular chasses evolved.

Membrane permeability negatively correlates with transporter gene repertoires
The observed differences in membrane permeability imply that any transition between archaeal and bacterialtype membrane chemistries would require extensive recalibration of numerous cellular systems in response to changes in permeability, osmotic stress, and metabolite homeostasis. Such a transition could be facilitated by a mixed archaeal-bacterial membrane [7,56], but the ultimate change would require adaptation to these altered cellular properties. The evolution of membrane transporters could permit a reduction in lipid membrane permeability, for example, if a cell was to transition from an archaeal-like membrane to a bacterial-like membrane, as predicted under some hypotheses for the origins of the eukaryotes [57,58]. Accordingly, given the increased permeability of the core archaeal lipid membranes shown here, we hypothesized that archaeal genomes would encode a significantly reduced complement of transporter gene families relative to Bacteria, particularly for those protein families known to transport metabolites capable of permeating archaeal lipid membranes (shown in Figure 2). A limited transporter repertoire could reflect a reduced dependency on protein-based translocation systems as metabolite requirements could be satisfied by a combination of core metabolic function and lipid membrane permeability. Likewise, increased membrane permeability may limit the utility of membrane transporters by decreasing transport efficiency or impairing the formation of concentration gradients.
To test this hypothesis, we iteratively searched diverse bacterial (n = 3,044) and archaeal (n = 243) genome-derived predicted proteomes using profile hidden Markov models (HMM, n = 277) derived from TCDB (Transporter Classification Database) protein families [59], to identify previously classified transporter homologs across prokaryotes. Despite the sensitivity of our search, the Archaea had fewer transporters relative to the Bacteria, irrespective of bacterial membrane system (e.g., monoderms or diderms) and regardless of whether transporter numbers were normalised to the total number of proteins per genome ( Figure   4A and Figure S7A). Certain transporter families were consistently encoded across prokaryotes, including many ion transporters (see cluster 2 in Figure 4A, largely composed of ion transporters), which are required due to the impermeability of both membrane types to ions, with the exception of protons [49]. In contrast, other families showed significantly reduced representation in Archaea (see clusters 1, 3, and 5 in Figure 4A which were functionally heterogeneous, whereas cluster 4 comprised outer membrane transporters associated with diderms). In particular, transporter families known to translocate metabolites similar to those that permeate archaeal membrane mimics (e.g., amino acids, sugars and nucleobases shown Figure 2) were significantly depleted even when accounting for differential taxon sampling bias using bootstrap resampling ( Figure 4B).
It has previously been argued that protein-membrane interactions, attuned to the specific lipid characteristics of either bacterial or archaeal membranes, may have acted to enforce the lipid divide as the Bacteria and Archaea diversified [14]. Therefore, to account for the possibility that archaeal transporters were not accurately recovered in our searches due to divergent biochemical characteristics or a lack of archaeal transporter family representation in TCDB, we first examined the possibility that archaeal membrane

CONCLUSIONS
The deepest branch in the tree of life separates the Archaea and the Bacteria. This ancient node demarks two different core membrane chemistries known as the lipid divide. There is considerable variation in membrane composition on both sides of the lipid divide, but fundamentally these membranes are built of two different phospholipid chemistries. Here we demonstrate that homogenous versions of the archaeal and bacterial core lipid membranes show distinct differences in permeability characteristics when generated using electroformation. Archaeal-type lipid membranes show permeability to a range of compounds that would theoretically be useful to known cellular metabolic networks. Our data demonstrate that archaeal membrane permeability is dependent on the simultaneous presence of methyl chain branching and ether bond properties, two hallmarks of archaeal lipids. This selective permeability could have provided early cellular forms with access to specific metabolic resources, from which a metabolic network could arise and diversify. Taken together with previous findings demonstrating that archaeal lipid membranes display low permeability to ions [16], these data support the suggestion that early cellular forms without a dedicated protein based transmembrane transportation system would be more likely to acquire stable and diversified metabolic functions using an archaeal lipid membrane than a bacterial lipid membrane [7].

Preparation of synthetic lipid vesicles
Briefly, the electroformation process was performed in three steps. The rise step: the AC voltage was linearly increased peak to peak (p-p) from 0 V till the maximum chosen value (see Table S3). The main step: the voltage was kept constant for the chosen duration (see Table S3). The fall step: the voltage was decreased linearly to 0 V. For lipids 9 and 10 none of the protocols employed yielded mechanically stable vesicles. For lipid 6 none of the employed protocols allowed vesicle electroformation possibly because this lipid forms nonlamellar structures of either a cubic or hexagonal fashion [48]. The protocols for the electroformation of vesicles made of lipids 7, 8 and 11 were adapted from [61]. The final fluorescent vesicle suspension consisted of fluorescent vesicles (because of the embedded CF molecules) and free CF molecules in the washing buffer.
This suspension was collected from the ITO-slide surface using a pipette, placed in an Eppendorf microcentrifuge tube for storage at 4 °C and used within five days.

Extraction of natural lipids
Lipids were extracted from the halophilic Haloferax volcanii (100 % diether lipids; [45]) and from the thermoacidophilic Sulfolobus acidocaldarius (ca. 10 and 90 % diether and tetraether lipids, respectively [62]). GmbH). We note that, compared to microfluidic devices previously employed for single-cell confinement [67,68], our approach relies on a single height device that can be fabricated in a single step without the use of dedicated photolithography equipment and can be easily carried out by users who might not be familiar with the technical aspects of microfabrication.

Microfluidic permeability assay
15 μL of washing buffer was injected in the microfluidic chip from the buffer inlet ( Figure 1)  intensity [69]. For each membrane mimic in Table S1, in order to account for the impact of both the delivery of the washing buffer solution and photobleaching on the intra-vesicle CF fluorescence signal, we performed separate control assays by connecting the metabolite inlet to a syringe containing the washing buffer solution instead of the metabolite solution. Apart from this modification, these control assays were carried out following the protocol described above for the microfluidic permeability assays. The intra-vesicle CF fluorescence consistently linearly decreased during the delivery of the washing buffer for all membrane mimics investigated and this information was used to provide a background signal for the corresponding microfluidic permeability assays (see Image and data analysis section below). To do so a correction factor was calculated from the microfluidic control assay data sets (see below), multiplied by each time value and added to the corresponding intra-vesicle fluorescence value (after the background and initial fluorescence value subtractions above). To obtain a correction factor for each membrane mimic, first we applied the image analysis protocol above to obtain the single-vesicle temporal dependence of intra-vesicle fluorescence values during the delivery of the washing buffer and subtract from these values the corresponding background and initial intra-vesicle fluorescence value. Second, we averaged these temporal dependences of corrected single-vesicle fluorescence values to obtain a mean temporal dependence for each membrane mimic during the delivery of the washing buffer. Finally, we fitted this mean temporal dependence to a linear regression with the intercept forced to zero and obtained the slope of the linear fluorescence decrease for each membrane mimic. These slope values were used as correction factors to calculate the permeability of each membrane mimic to each metabolite as described above. For some of the membrane mimic and metabolite pairs we noticed a minority of outliers (i.e. vesicles that became either much brighter or dimmer with respect to the majority of vesicles investigated during metabolite delivery, see for example Figure 2B).

Image and data analysis
This heterogeneity is common when investigating membrane permeability both in vitro [40] and in vivo [39,43].

In silico identification of prokaryotic transporter proteins
To characterize the transporter repertoires of bacterial and archaeal species in silico, we first assembled a collection of prokaryotic reference proteomes from UniProt (release 2021_03 [59]). To minimize taxonomic redundancy, a single proteome was selected per genus based on BUSCO completeness scores [71], resulting in 3,044 bacterial and 244 archaeal proteomes. Metagenomes from the bacterial candidate phyla radiation (CPR) were excluded due to their prevalence and lack of morphological information. To comprehensively identify transporter homologs, profile hidden Markov models (HMMs) derived from TCDB (Transporter Classification Database) protein families (termed tcDoms, downloaded 2 June 2021) were used to search each proteome using HMMER v3.1b2 (E < 10 -5 , incE < 10 -5 , domE < 10 -5 ) [59,72]. If multiple HMMs identified the same predicted protein, the protein was assigned to the family with the lower E-value. To improve the sensitivity of the HMMs, the hits from the initial HMM search were aligned using MAFFT v7.471 (-auto), trimmed with a gap-threshold of 10% using trimAl v1.4, and the resulting alignments were used to generate new HMMs [59,73]. Using the second iteration HMMs, another search was conducted as above, producing the final set of identified proteins. To reduce the potential effects of taxon sampling and dataset bias, prokaryotic taxa were grouped into their respective orders based on NCBI Taxonomy classifications [74].
Transporter abundance was then interpreted as the median number of transporters assigned to a given TCDB family, normalized by the total number of proteins encoded by each taxa, across each order. The resulting distribution was visualized in R v4.0.2 and hierarchical clustering was done using euclidean distances and the Ward.D2 clustering method implemented by pheatmap (https://mran.microsoft.com/snapshot/2018-08-31/web/packages/pheatmap/pheatmap.pdf). The differential abundance of individual transporter families was assessed by comparing archaeal and monoderm transporter abundances (given their morphological similarities) using Wilcoxon tests after Bonferroni correction.
To identify putative archaeal transporter families undetected by the TCDB HMMs, we clustered archaeal proteomes into protein families using the Markov clustering algorithm (I=1.4) based on pairwise BLASTp searches conducted using Diamond v2.0.9.147 (E < 10 -5 , query coverage > 50%, -max-target-seqs =10 5 , --more-sensitive search option) [75,76]. Transmembrane (TM) domains were then predicted for each archaeal protein using Phobius and the median number of TM domains was determined for each protein family [77]. Those families with representation in at least five archaeal species and with a median of at least four TM domains were identified and annotated using eggNOG mapper v2.1.2 [78]. TM domain-containing protein families were classified as putative transporters if they were annotated with PFAM domains associated with transporter function (e.g., all transporter and channel-associated domains) [79].    Table S1. However, care has been taken to obtain the same N for each metabolite-experiment across each pair of lipid type to ensure reliable statistical comparisons. The lipids used for creating archaeal 4ME diether G1PC, bacterial diester G3PE-PG-CA, 4ME diether G3PC, 16:0 Diether G3PC, or 4ME 16:0 Diester G3PC vesicles are lipids 1, 2, 3, 4 and 5, respectively, in Table S1.    Table S1.  Table S1) because the transition temperature of these lipids (i.e. 24 °C) is very close to room temperature and vesicles easily burst during our permeability assays. Finally, we could not form vesicles using lipids with a chain length of 6 carbons (lipid 10 in Table S1) despite attempting different electroformation protocols (Table   S4). The lipids used for creating vesicles with tail length of 12, 16 and 18 carbons are lipids 8, 4 and 9, respectively, in Table S1. creating the archaeal membrane mimics with and without saturation are lipids 9 and 12, respectively, in Table   S1.   Table S3. Protocols to obtain vesicles using synthetic lipids. Parameters for the protocols employed in this work. The protocols reported in grey yielded a negative outcome (see Methods).

Supplementary file 1. Statistical comparisons between permeabilities of two different lipid vesicle
types. Summary of significance and p values estimated using two-tailed Mann-Whitney tests between distributions of CF fluorescence after 3 min of delivery of glycine, alanine, leucine, aspartic acid, glutamine, tryptophan, glyceraldehyde, dihydroxyacetone, glycerol, deoxyribose, ribose, arabinose, urea, cytosine, uracil, phosphonate, adenine or guanine to individual vesicles made of archaeal 4ME diether G1PC or bacterial diester G3PE-PG-CA lipids.
Supplementary file 2. Statistical comparisons between permeabilities of more than two different lipid vesicle types. Mean rank difference, summary of significance and p-value estimated using the Kruskal-Wallis one-way analysis of variance test between distributions of CF fluorescence after 3 min of delivery of urea, glycine, ribose, deoxyribose, glycerol and phosphonate to individual vesicles made of archaeal 4ME diether